Shyam Mukherjee

Benchmarked multiple text-to-SQL models to evaluate their performance across various datasets, aiming to identify the most effective solutions for accurate query generation.

I focused on refining prompts to enhance the precision and optimization of SQL queries, leading to improved data retrieval. I also developed reusable query templates by abstracting and substituting literals, which streamlined the process and significantly reduced the number of API calls, thus lowering associated LLM usage costs. The project also included fine-tuning LLMs on specific domain data to improve their contextual understanding and query accuracy, resulting in a more efficient and intelligent database querying system.

Developed a specialized web crawler for domain-specific content retrieval by integrating Apache Nutch, Solr, and Lucene. The system effectively identified the domains of the crawled documents, ensuring targeted and relevant data collection. It utilized Wikipedia category graphs to accurately tag document topics, enhancing the precision of content classification and retrieval. The project involved configuring and optimizing Apache Nutch for scalable crawling, indexing the content with Solr for efficient search and retrieval, and leveraging Lucene's powerful indexing and search capabilities. This comprehensive approach facilitated the extraction of high-quality, domain-specific information for improved data analysis and research.